A high-quality evaluation of an innovation or intervention in healthcare will typically use quantitative and qualitative evidence.
The latter can reveal context and an understanding of factors influencing success or failure. Analysis of qualitative evidence, however, is a very time-consuming process and can be subject to problems of reviewer fatigue and error.
To address this, we have developed and trialled new approaches to analysis, using large language models (LLMs) to increase the robustness of analyses and reduce resource use. The trials have demonstrated significant savings in reviewer time and have opened up the possibility of adopting more flexible and responsive approaches to analysis. These techniques have been used in a number of real-world evaluations and we continue to develop and refine them in an increasing range of use cases.
